[llvm] SLP: Remove artificial shifts from min-bitwidth test (NFC) (PR #222634)

[Hari Limaye via llvm-commits]

================

----------------
hazzlim wrote:

The issue is that I’m specifically trying to avoid foldable shifts here, as a follow-up will teach the AArch64 cost model that these shifts fold into the logical instructions. Using an argument as the shifted value would still be recognized as folded by the follow-up cost-model change. We could use that form, but the follow-up would then need to lower -slp-threshold further to preserve the test coverage. Would you prefer that over making this test independent of folded-shift costing?

https://github.com/llvm/llvm-project/pull/222634